Product Description

It can be used to detect the quality of rice in food processing plants, government grain storage warehouses and grain quality inspection centers. The equipment uses the latest AI vision detection technology and is equipped with 3 high-resolution cameras to analyze the attributes of the front and back sides of the rice. The rice on the front and back is registered one by one, and combined with their respective attributes to synthesize the attributes of a complete rice; the deep neural network is used to segment the attached rice at the instance level to easily deal with the situation of rice adhesion; at the same time, the cloud platform is opened and the samples of different customers can be remotely trained to meet customer customized classification standards.

Inspection principle

Manual sampling, inspection, recording, and statistics have disadvantages such as slow speed, low accuracy, high missed and false positive rates, and long-term fatigue. This machine can replace manual work, can work 7*24 hours, detect the quality of rice with high precision, detect broken rice, chalky rice, imperfect rice, and moisture in the rice in time, and find whether there are mildew, worms, impurities and other problems. It can be used for daily sampling inspection before and after rice production.

The rice quality detector can be connected to upstream and downstream production equipment according to the specific production needs of customers on site. The parts in contact with the equipment and samples are made of medical-grade materials. It is safe and hygienic, with intelligent design, simple operation and convenient maintenance.

Testing system display:

Key technology

Combine traditional machine vision methods and artificial intelligence algorithms to analyze rice. First, use traditional vision methods to segment the rice grains in the video frame, and then use artificial intelligence algorithms to identify the attributes of the segmented rice grains to determine whether there are insects, moth, sprouting, mildew and other problems. At the same time, two high-resolution cameras were used to photograph the front and back of the rice, and the properties of the two sides were analyzed. Through the registration algorithm, the front and back of the rice are registered one by one, and their respective attributes are combined to obtain the attributes of a complete rice grain.

1. Automatic binarization: Use deep neural network to segment the foreground and background of the image. Compared with the traditional binarization method, it can be applied to a variety of lighting conditions, and the edge segmentation of rice is smoother, fast and robust High advantages.

2. Adhesive rice segmentation algorithm: The method based on connected domains cannot segment the adhered rice. The deep neural network is used to segment the adhered rice at an instance level, which can reach a speed of 1000fps and can process the adhered rice in real time.

3. Rice attribute recognition algorithm: adopts a lightweight neural network and integrates a semi-supervised learning method. The model can be iteratively optimized only by marking a small amount of data. It has the advantages of high accuracy, fast speed, and convenient deployment.